real-stanford / diffusion_policy

[RSS 2023] Diffusion Policy Visuomotor Policy Learning via Action Diffusion
https://diffusion-policy.cs.columbia.edu/
MIT License
1.1k stars 206 forks source link

fixed upsampling so that h_local[1] is actually used #10

Closed Aaditya-Prasad closed 10 months ago

Aaditya-Prasad commented 1 year ago

Previous upsampling if statement on line 229 of conditional_unet1d.py never triggers since idx is 0 indexed. I assume this is a typo and that expected behaviour is for h_local[1] to be added to x on the last of the upsampling layers.

cheng-chi commented 10 months ago

Hi @Aaditya-Prasad this is indeed a bug. However, this change will break compatibility with existing checkpoints. Therefore I will add your suggested change as a comment in the code. Thanks!